People measuring RPS (requests per second) are rarely measuring CPS (connections per second), so the TLS setup overhead would be amortized across numerous requests.
A single webpage will often load a number of asset files over the same underlying connection, so it's not like CPS is the single, entirely accurate measure and RPS is just bad or something... they're both valid, interesting numbers.
Even then, TLS setup isn't that expensive, in my opinion. Someone else here mentioned 35% overhead from TLS on a Raspberry Pi, which seems high to me... but it would still be orders of magnitude better performance than what the original article on this thread was claiming.
